Comments to date on reaffirmation of AES-4id-2001 AES information document for room acoustics and sound reinforcement systems -- Characterization and measurement of surface scattering uniformity, published 2007-02-13 for comment.
The present notes come from the experiments that followed the AES recommendation document on reaffirmation and that were carried out by our team in Parma in year 2006 on the floor of a large industrial shed. They were described in the paper: "AES 121 Convention Paper 6922 - Surface scattering uniformity measurements in reflection free environments". Some of the more recent results are to be published at ICA2007 in Madrid in September (a preview of the paper can be given to the evaluating committee if requested) and will be presented also to 124th AES convention in New York.
We consider the boundary measurement to be more specific for low-cost and preliminary studies and we would like to share our experience to expand the document informations on this type of measurement. Also we point to some discrepancies we noted in executing the measurements.
Here are the modifications we suggest and the parts of the existing document that we propose to give more attention to:
- Page 4, Chapter 3.1, second line states the reference flat surface should be 'thin', we propose it should be changed into 'as thick as the panel under test'.
Note: we often used the panels' flat back faces as references, these guarantee to measure a well definite border effect and low frequency diffraction that are better related to the panels under study, hence the comparison of results permits to define the effect of scattering more clearly.
- Page 7, Chapter 3.16 in the Note, the required 0.1 difference in the diffusion uniformity coefficient can be insufficient to define a diffusion bandwidth: this actually depends on the previous point 3.1 Note: this is linked to the choice of the reference surface and it actually depends on the definition of the diffusion phenomenon. The measures on the panel's back flat faces show a first maximum defining a monopole-like diffraction, and a second maximum that is strongly related to the border effect. The panel's front face show a very different behaviour and the 0.1 limit can actually be misleading.
- Page 8, chapter 4.1 (before figure 5) states: "Boundary measurements may also be carried out to remove the necessity for a space to be anechoic in one plane, provided conditions in annex A are satisfied", here it should be added: "hence diffusion analysis is possible only on two orthogonal planes and the measurements are specific for single-plane diffusers".
Note: this is just for clarifying the concept.
- Page 10, chapter 188.8.131.52 we suggest figure 7 should show more receiver angular resolutions to clearly prove the decision on a 5-degree angle.
Note: some tests of ours on 7.5 degree vs. 3.75 degree angles calculations show little difference in the values of the coefficient d obtained (see cited AES convention paper #6922). The angular resolution is important because it is related to the number of channels and measurement microphones needed for this kind of measurements (read an economic restraint).
- Page 10 chapter 184.108.40.206 third line, after "time is limited by the conditions of measurement" could be stated in parenthesis "i.e. manually operated measurements".
Note: such as the cases in less expensive non anechoic environments.
- Page 10 chapter 220.127.116.11 in the last sentence 5 angles of sound incidence might be better that the 3 suggested ones for a fast measure.
Note: Some preliminary studies of ours show that in this case 5 incidence angles (-60 degrees, -30 degrees, 0 degrees, 30 degrees, 60 degrees) give a good estimate of the average on more incidence angles, better than the proposed 3 angles.
- Page 11 chapter 4.4.1. the last line should be changed into "... time-delay spectrometry (TDS), maximum-length sequences (MLS) and logarithmic sine sweep technique. Of these methods logarithmic swept sine is the fastest and most repeatable method."
Note: in our studies we proved to have saved 60% of operation time, elaboration time is saved as well since just one convolution is needed with no averaging on multiple measures, using generally much shorter signals. System non linearities are completely rejected.
- Page 13 Figure 10 should specify the name of the signals in the various graph boxes, in order: h3,i (t) ; h2,i (t); h1,i (t); h2,i (t) - h1,i (t); h4,i (t) ; h4w,i (t) it should also specify the axis unit of measure.
- Page 14 Figure 11 should have an explicit legend. For example "1. N-channel measurement set up plan", "2. Single h1(t) impulse response measure", "3. Series of N deconvolved h4(t) reflections as required", "4. N frequency responses H4(f) before filtering", "5. Third-octave-band diffusion polar graphs after required filtering", "6. Third octave diffusion coefficient graph".
Note: We suggest also a sonar-like representations of h1, h2, h4 (with all the single channels put side by side: axis x for the microphone number, axis y for time and a colour scale for the amplitude in dB) as intermediate useful verification steps in the data processing protocol. They could actually substitute graph 2 and 3 in figure 11 showing in 2 dimension graphs what is now shown with 1 dimensional ones.
- Page 15, chapter 4.5.3 first sentence: we suggest to require a smoothed-edges time window instead of the rectangular one.
Note: rectangular time windows often generates strong oscillations at high frequencies, the Adrienne window in EN 1793-5 could be considered, with a shorter roll-on and roll-off period.
- Page 15 chapter 4.5.3 at the seventh line should be added "in any case the window size shall be longer than a minimum 10 milliseconds for the standard measurement geometry (r1 = 5 m; r2 = 10 m)".
- Page 15 chapter 4.5.4 we suggest to cancel the last phrase on zero padding but, as already stated, specify in point 4.5.3 a minimum window length.
Note: The "three one-spectral points in all of the third octave frequency bands requirement" together with the "no zero padding" requirement one actually give a strong limitation to the temporal window length. This can demonstrate to be too severe in boundary non-anechoic measurements where window length is a key factor in deciding the experiment's location from its size.
So we suggest the document to be more explicit, one solution would be specifying a minimum window dimension, from our experiments a 10 msec window is sufficient to study most of the first reflections from the panels. Zero padding should be permitted if the time window's edges are smoothed.
Example: 3 points in the 50 Hz third octave bands require a maximum spectral resolution of about 3.83 Hz, if fs = 48 KHz this translates in requiring a time window that is longer than 12522 samples, 2^14 will be the best choice for dsp related velocity reasons, this means a secondary reflection free 341 milliseconds long time window (117 meters of travel distance, about 58.5 m from the first important reflective object or surface) which can demonstrate to be too much in reasonable industrial shed dimensions, even in the case the first second order reflections still permit to respect the 40 dB SNR requirement in chapter 4.5.3.
- Page 19 - Annex A could be more detailed on non-anechoic measurement, here we advise a note for the acquisition operating protocol in boundary measurements. A further chapter should be added to the annex: "A.2.4 Acquisition. When numerous panels are to be compared, all of the panels' h1 with the same sound incidence angle should be acquired within a short time period, comprising the acquisition of the respecting h2."
Note: This guarantees maximum correlation between subsequent impulse response measurements.
Angelo Farina (Member), Dipartimento di Ingegneria Industriale, Universita di Parma, Italy (firstname.lastname@example.org)
Lorenzo Rizzi (Associate Member), Gruppo LAE, Parma, Italy (email@example.com)
I must apologise for the delay in responding to your comment of 2007-04-23, due in part to the AES 122nd Convention in Vienna.
We are grateful for your proposals to revise this standard. At this stage, however, the standard must be reaffirmed or it will lapse because it is overdue its 5-year mandatory review period.
However, these notes will be a valuable contribution to a subsequent revision of AES-4id-2001 (r2007).
Mark Yonge AES Standards Secretary